Optimization of Large Determinant Expansions in Quantum Monte Carlo (original) (raw)
Related papers
Journal of Chemical Theory and Computation
We present an improved formalism for quantum Monte Carlo calculations of energy derivatives and properties (e.g. the interatomic forces), with a multideterminant Jastrow-Slater function. As a function of the number N e of Slater determinants, the numerical scaling of O(N e) per derivative we have recently reported is here lowered to O(N e) for the entire set of derivatives. As a function of the number of electrons N , the scaling to optimize the wave function and the geometry of a molecular system is lowered to O(N 3) + O(N N e), the same as computing the energy alone in the sampling process. The scaling is demonstrated on linear polyenes up to C 60 H 62 and the efficiency of the method is illustrated with the structural optimization of butadiene and octate-1
Quantum Monte Carlo with very large multideterminant wavefunctions
Journal of computational chemistry, 2016
An algorithm to compute efficiently the first two derivatives of (very) large multideterminant wavefunctions for quantum Monte Carlo calculations is presented. The calculation of determinants and their derivatives is performed using the Sherman-Morrison formula for updating the inverse Slater matrix. An improved implementation based on the reduction of the number of column substitutions and on a very efficient implementation of the calculation of the scalar products involved is presented. It is emphasized that multideterminant expansions contain in general a large number of identical spin-specific determinants: for typical configuration interaction-type wavefunctions the number of unique spin-specific determinants Ndetσ ( σ=↑,↓) with a non-negligible weight in the expansion is of order O(Ndet). We show that a careful implementation of the calculation of the Ndet -dependent contributions can make this step negligible enough so that in practice the algorithm scales as the total number...
The Journal of chemical physics, 2016
The Full Configuration Interaction Quantum Monte Carlo (FCIQMC) method has proved able to provide near-exact solutions to the electronic Schrödinger equation within a finite orbital basis set, without relying on an expansion about a reference state. However, a drawback to the approach is that being based on an expansion of Slater determinants, the FCIQMC method suffers from a basis set incompleteness error that decays very slowly with the size of the employed single particle basis. The FCIQMC results obtained in a small basis set can be improved significantly with explicitly correlated techniques. Here, we present a study that assesses and compares two contrasting "universal" explicitly correlated approaches that fit into the FCIQMC framework: the [2]R12 method of Kong and Valeev [J. Chem. Phys. 135, 214105 (2011)] and the explicitly correlated canonical transcorrelation approach of Yanai and Shiozaki [J. Chem. Phys. 136, 084107 (2012)]. The former is an a posteriori inter...
Recent advances in determinant quantum Monte Carlo
Philosophical Magazine, 2013
Determinant quantum Monte Carlo is a method for studying magnetic, transport and thermodynamic properties of interacting fermions on a lattice. It is widely used to explore the physics of strongly correlated quantum systems, from cuprate superconductors to ultracold atoms trapped on optical lattices. This paper contains a description of recent algorithmic advances in the determinant quantum Monte Carlo technique. Focus will be on algorithms developed for hybrid multicore processor and GPU platforms. The resulting speed-up of the simulations will be quantified. Simulations' results will also be presented, with an emphasis on physical quantities that can now be computed for large numbers of sites.
Simple formalism for efficient derivatives and multi-determinant expansions in quantum Monte Carlo
The Journal of Chemical Physics, 2016
We present a simple and general formalism to compute efficiently the derivatives of a multi-determinant Jastrow-Slater wave function, the local energy, the interatomic forces, and similar quantities needed in quantum Monte Carlo. Through a straightforward manipulation of matrices evaluated on the occupied and virtual orbitals, we obtain an efficiency equivalent to algorithmic differentiation in the computation of the interatomic forces and the optimization of the orbital paramaters. Furthermore, for a large multi-determinant expansion, the significant computational gain recently reported for the calculation of the wave function is here improved and extended to all local properties in both all-electron and pseudopotential calculations.
Journal of Chemical Theory and Computation, 2018
We investigate the performance of a class of compact and systematically improvable Jastrow−Slater wave functions for the efficient and accurate computation of structural properties, where the determinantal component is expanded with a perturbatively selected configuration interaction scheme (CIPSI). We concurrently optimize the molecular ground-state geometry and full wave functionJastrow factor, orbitals, and configuration interaction coefficientsin variational Monte Carlo (VMC) for the prototypical case of 1,3-transbutadiene, a small yet theoretically challenging π-conjugated system. We find that the CIPSI selection outperforms the conventional scheme of correlating orbitals within active spaces chosen by chemical intuition: it gives significantly better variational and diffusion Monte Carlo energies for all but the smallest expansions, and much smoother convergence of the geometry with the number of determinants. In particular, the optimal bond lengths and bond-length alternation of butadiene are converged to better than 1 mÅ with just a few thousand determinants, to values very close to the corresponding CCSD(T) results. The combination of CIPSI expansion and VMC optimization represents an affordable tool for the determination of accurate ground-state geometries in quantum Monte Carlo.
Semi-stochastic full configuration interaction quantum Monte Carlo: Developments and application
The Journal of chemical physics, 2015
We expand upon the recent semi-stochastic adaptation to full configuration interaction quantum Monte Carlo (FCIQMC). We present an alternate method for generating the deterministic space without a priori knowledge of the wave function and present stochastic efficiencies for a variety of both molecular and lattice systems. The algorithmic details of an efficient semi-stochastic implementation are presented, with particular consideration given to the effect that the adaptation has on parallel performance in FCIQMC. We further demonstrate the benefit for calculation of reduced density matrices in FCIQMC through replica sampling, where the semi-stochastic adaptation seems to have even larger efficiency gains. We then combine these ideas to produce explicitly correlated corrected FCIQMC energies for the beryllium dimer, for which stochastic errors on the order of wavenumber accuracy are achievable.
Optimizing large parameter sets in variational quantum Monte Carlo
Physical Review B, 2012
We present a technique for optimizing hundreds of thousands of variational parameters in variational quantum Monte Carlo. By introducing iterative Krylov subspace solvers and by multiplying by the Hamiltonian and overlap matrices as they are sampled, we remove the need to construct and store these matrices and thus bypass the most expensive steps of the stochastic reconfiguration and linear method optimization techniques. We demonstrate the effectiveness of this approach by using stochastic reconfiguration to optimize a correlator product state wavefunction with a pfaffian reference for four example systems. In two examples on the two dimensional Hubbard model, we study 16 and 64 site lattices, recovering energies accurate to 1% in the smaller lattice and predicting particle-hole phase separation in the larger. In two examples involving an ab initio Hamiltonian, we investigate the potential energy curve of a symmetrically dissociated 4x4 hydrogen lattice as well as the singlet-triplet gap in free base porphin. In the hydrogen system we recover 98% or more of the correlation energy at all geometries, while for porphin we compute the gap in a 24 orbital active space to within 0.02eV of the exact result. The numbers of variational parameters in these examples range from 4 × 10 3 to 5 × 10 5 , demonstrating an ability to go far beyond the reach of previous formulations of stochastic reconfiguration.